Refactor: Improve the buffer close for vision_llm_chunk (#9845)

### What problem does this PR solve?

Improve the buffer close for vision_llm_chunk

### Type of change

- [x] Refactoring

rag/app/picture.py

@@ -78,14 +78,11 @@ def vision_llm_chunk(binary, vision_model, prompt=None, callback=None):
     txt = ""
 
     try:
-        img_binary = io.BytesIO()
+        with io.BytesIO() as img_binary:
             img.save(img_binary, format='JPEG')
             img_binary.seek(0)
-
             ans = clean_markdown_block(vision_model.describe_with_prompt(img_binary.read(), prompt))
-
             txt += "\n" + ans
-
             return txt
 
     except Exception as e: